A social network for machines-realizing industry 4.0

Abstract

The technology developed in this project is radically essential, globally sought-after and involves multiple every engineering domain. The project is on Industry 4.0, or what is considered to be the next Industrial Revolution – the application of Cyber-Physical Systems (CPS), Industrial IoT and Computer Optimization Techniques in manufacturing enterprises to create a system capable of remote monitoring, intelligent planning and autonomous decision making. At its most fundamental level, it enables ‘non-human entities to autonomously interact with each other and with humans, to intelligently work towards their goals, and make data-centric decisions using cyberphysical transformations and Internet of Things.’Due to the huge diversity of industrial assets in terms of structure, operation and automation level, in addition to the shallow penetration of embeddedsensor technology into the existing systems, existing development model do not suffice the adequacy of Industry 4.0, thus making deployment of such CPS a bottle-neck. To overcome this bottle-neck, an event-based approach is developed for creation of a transient system, ‘cyber-twin,’ for industrial assets expediting the development of CPS for Industry 4.0 realization. A cyber-twin is a virtual replica of the physical asset, have the capability to simulate the asset’s behavior and forecast the state of the asset in future. It is intelligent - capable of making decisions autonomously, and social - capable of interacting with other cyber-twins in the enterprise. An event-based cyber-twin for a semi-automatic special-purpose facing and centering machine at a manufacturing facility is developed here. Advantageous characteristics of an event-based cyber-twin include: i) developmental suitability irrespective of structure, operation or automation level; ii) induced connectability of the machines and other departments in the industrial network; iii) upgradability of hardware (though more sensors/embedded sensors) and performance; ivintegrability into the physical-machine; v) analytics and operations planning and vi) employability of established standards such as MTConnect.Since a significant aspect of developing such intelligent cyber-twins is its ‘adaptive’ nature i.e. the capability to incorporate real-time actions (new data sets) into functioning, a case study is further presented to implement a Bayesian algorithm to update the time to failure distribution parameters for a machine tool with the occurrence of every new failure event. The study draws some light on the following two inter-related points. (i) when to update the prediction parameters (i.e. after how many new data points) and (ii) effect of presence of multiple failure modes on Bayesian results. The work further develops novel algorithms for operations planning using the cyber-twins. In this work, examples of planning and optimization of preventive maintenance and production schedule are considered.Optimizing the preventive maintenance schedule is important to minimize operations cost and machine downtime. Moreover, for a multi-component machine, the knowledge of which components to perform preventive maintenance on, can be crucial for such optimization. A stochastic simulation model can be used to evaluate all possible candidates and find the optimum preventive maintenance schedule. However, this involves infeasible computational time owing to the combinatorial nature of the problem. A Memetic Algorithm is proposed as a heuristic in the present work to address this challenge. Accuracy of obtained solutions and run-time is compared with brute-force search method and genetic algorithm for the same system. The Memetic Algorithm is found to yield better results as it can explore the search space more efficiently.Further, as these advancements endow intelligence to every individual entity on the shop floor, it becomes necessary to develop new schemes that can unlock the potential of decentralized data observation and decision-making, enabled through an Industrial Internet of Things. A distributed algorithm is developed for the first time that performs intelligent maintenance planning for identical parallel multi-component machines in a job-shop manufacturing scenario. The results are breakthrough in terms of computation time, quality of solution and scalability. The supremacy of the devised algorithm is demonstrated over conventional centralized heuristics such as Memetic Algorithm and Particle Swarm Optimization. We believe that such distributed algorithms will give new research directions to conventionally done centralized operations planning approaches in the industry. Further, it becomes an integral and a very important part for any Industry 4.0 implementation.